Bring vidya in

egui is the only backend anyone wants now, so the reason libvidya's symbols
were kept interchangeable — a raylib build, a cimgui build, swapped by
search path alone — was paying for a choice nobody is going to make. With
that gone there is no argument left for vidya being somewhere else.

Raw, without history: crates/vidya-core is the egui layer,
crates/jolt-vidya the C ABI over it, jolt/glimmer-vidya the jolt side. The
raylib and cimgui backends, the buck2 vendoring and the Android host did
not come, and neither did 6.5G of build output.

Both objects now come out of one target directory, so a consumer points
LD_LIBRARY_PATH at one place and still links only what it names.

opt-level 2 moves to the workspace root. vidya-ffi set it when it was its
own workspace, and a member's profile is ignored — the tree walk and the
texture uploads are hot every frame.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>